@remnic/bench 9.69.86 → 9.69.87

package/dist/{chunk-P4VDWOOT.js → chunk-LJCLTHYJ.js}

@@ -4989,6 +4989,21 @@ var H6_FROZEN_SPLITS = Object.freeze({
     "h6-task-30"
   ])
 });
+var H6_FROZEN_EVIDENCE_PATHS = Object.freeze([
+  "decision-rule-timing.json",
+  "trap-audit.json",
+  "tasks/task_01_cache_invalidation",
+  "tasks/task_02_async_queue",
+  "tasks/task_03_token_bucket",
+  "tasks/task_04_event_emitter",
+  "tasks/task_05_trie_search",
+  "tasks/task_06_connection_pool"
+]);
+function isFrozenEvidencePath(fixturePath) {
+  return H6_FROZEN_EVIDENCE_PATHS.some(
+    (evidence) => fixturePath === evidence || fixturePath.startsWith(`${evidence}/`)
+  );
+}
 var TrapFingerprintV1Schema = z2.object({
   version: z2.literal(1),
   trapId: z2.enum(H6_TRAP_IDS),
@@ -8417,7 +8432,7 @@ async function generateH6BenchmarkDataset(seed = 81) {
 // src/coding-graph/repo-gen/validator.ts
 import { execFileSync as execFileSync3 } from "child_process";
 import { createHash as createHash7, randomUUID } from "crypto";
-import { cp, lstat as lstat2, mkdir as mkdir6, mkdtemp as mkdtemp2, readFile as readFile5, readdir as readdir3, rename as rename2, rm, writeFile as writeFile4 } from "fs/promises";
+import { copyFile, cp, lstat as lstat2, mkdir as mkdir6, mkdtemp as mkdtemp2, readFile as readFile5, readdir as readdir3, rename as rename2, rm, writeFile as writeFile4 } from "fs/promises";
 import { basename, dirname as dirname2, join as join2, parse, resolve, sep } from "path";
 
 // src/coding-graph/repo-gen/import-scanner.ts
@@ -9393,6 +9408,24 @@ async function writeH6FixtureBundle(outputDir, input) {
           force: false
         });
       }
+      for (const evidence of H6_FROZEN_EVIDENCE_PATHS) {
+        const evidenceSource = join2(absoluteOutputDir, evidence);
+        const evidenceDetails = await optionalLstat(evidenceSource);
+        if (!evidenceDetails) continue;
+        const evidenceTarget = join2(stagingDir, evidence);
+        if (evidenceDetails.isDirectory()) {
+          await cp(evidenceSource, evidenceTarget, {
+            recursive: true,
+            errorOnExist: true,
+            force: false
+          });
+        } else if (evidenceDetails.isFile()) {
+          await mkdir6(dirname2(evidenceTarget), { recursive: true });
+          await copyFile(evidenceSource, evidenceTarget);
+        } else {
+          throw new Error(`Frozen evidence artifact is not a regular file or directory: ${evidence}`);
+        }
+      }
       backupDir = join2(
         dirname2(absoluteOutputDir),
         `.${basename(absoluteOutputDir)}.backup-${randomUUID()}`
@@ -17441,6 +17474,8 @@ export {
   H6_FROZEN_SEED,
   H6_FROZEN_INVENTORY_HASH,
   H6_FROZEN_SPLITS,
+  H6_FROZEN_EVIDENCE_PATHS,
+  isFrozenEvidencePath,
   TrapFingerprintV1Schema,
   ActionIntentV1Schema,
   SyntheticFileSchema,

package/dist/h6-timing-rerun.md

@@ -0,0 +1,154 @@
+# H6 timing rerun preregistration
+
+Status: fixed before any rerun row executes. This is the second registered run
+of the H6 failure-gate study. It decides one hypothesis, H6-timing, on new
+data. The machine-readable rule is
+`packages/bench/fixtures/h6-failure-gate/decision-rule-timing.json`, decision
+rule version 13. The rule records this file's raw SHA-256 hash; the runner
+verifies that hash before writing any run artifact. This document defines the
+design and how to decide the result. It does not state a result.
+
+## Why a second run exists
+
+The first main run (decision rule version 12) completed all 1,890 rows. Its
+timing comparison produced a repeated-failure benefit of 38.04 points on the
+17 complete tasks, but one pre-action row on `h6-task-10` was invalid with
+`VAGUE_CHECK` after a token-cap termination. Version 12 allowed zero primary
+task cuts, so H6-timing was `NOT_ESTIMABLE` by rule. The first run's retry
+rule forbids rerunning any returned task result, so a confirmatory answer
+requires a new manifest and new rows. That is this run.
+
+The content hypothesis is not part of this run. Version 12 decided it:
+`REJECTED`. The no-trap check is not part of this run. Version 12 measured it
+on 540 no-trap rows; the steps interval sat inside its margin and the
+pass-rate question was recorded as unresolved by that design. Neither question
+is reopened here, and this run cannot change either answer.
+
+## The question
+
+**H6-timing:** Does a fixed failure note work better just before a matched
+action than at turn start?
+
+## Design
+
+- 18 main-split tasks, the same frozen split as version 12
+- 3 run-time variants per task
+- 2 arms: `TURN_START_FAILURE` and `PRE_ACTION_FAILURE`
+- 5 seeds: **6, 7, 8, 9, 10**
+- 1 model profile: `h6-qwen35-35b-32k-nothink-q4km-v4`, the same
+  `(modelProfileId, modelProfileHash)` pair as version 12
+- no no-trap rows
+
+Expected rows: 18 x 3 x 5 x 2 = **540**. The manifest lists every expected row
+before execution.
+
+The seeds are new. Seeds 1 through 5 produced the first run's episodes; reusing
+them would remeasure the same trajectories under rules chosen after those
+trajectories were seen. Seeds 6 through 10 produce episodes that do not exist
+when this rule freezes.
+
+Everything else carries over from version 12 unchanged: the dataset inventory
+hash, task revisions, variant set, arm injection contracts for the timing
+pair, prompt contract, tool locks, sandbox flags, caps (including the 20,480
+cumulative token cap from Amendment 2), the retry rule (five host-fault
+retries, then pause; never rerun a returned task result), statistics seed 81,
+and 10,000 bootstrap and shuffle draws.
+
+## Outcomes and analysis
+
+Row outcomes, task-level averaging, the bootstrap, and the shuffle test are
+unchanged from version 12. The estimands are unchanged:
+
+- Repeated-failure benefit: `risk(TURN_START_FAILURE) - risk(PRE_ACTION_FAILURE)`
+- Relative risk reduction: the benefit divided by `risk(TURN_START_FAILURE)`
+
+The hypothesis family has exactly one member, so the Holm-adjusted p value
+equals the raw p value. The support rule is unchanged from version 12:
+H6-timing is `SUPPORTED` only if the absolute repeated-failure benefit is at
+least 0.05, the point relative risk reduction is at least 0.30, the 95%
+benefit interval's lower bound is strictly above zero, and the adjusted p
+value is strictly below 0.05. An estimable result that misses any condition is
+`REJECTED`. A zero observed baseline risk makes the ratio undefined and the
+result `NOT_ESTIMABLE`.
+
+The study decision maps directly: `SUPPORTED` gives `PASS`, `REJECTED` gives
+`REJECT`, `NOT_ESTIMABLE` stays `NOT_ESTIMABLE`.
+
+## Invalid rows
+
+This is the one scoring change from version 12, and it is fixed here before
+any row runs.
+
+A row invalid with `VAGUE_CHECK` — the offline check is missing, returns
+`INDETERMINATE`, or cannot prove the registered failure class — is **scored,
+not cut**. It takes the worst value for the pre-action arm on every metric:
+
+- in `PRE_ACTION_FAILURE`: `repeatedFailure = true`, `taskPassed = false`
+- in `TURN_START_FAILURE`: `repeatedFailure = false`, `taskPassed = true`
+- `steps = 0` in either arm; steps feed no timing decision
+
+Every imputed row is listed in `statistics.json` with its task, arm, seed, and
+variant. This scoring can only shrink the measured benefit and the relative
+risk reduction, so it cannot manufacture support, and it removes the failure
+mode that voided the first run: a cap-terminated episode whose check cannot
+classify the final repository state. The `taskPassed` imputation points the
+same direction. The pre-action arm records a failure and the turn-start arm
+records a pass, so the reported task-pass benefit can only move against the
+pre-action arm.
+
+Every other invalid reason — `START_DRIFT`, `TRACE_GAP`, `MIXED_ARM_STATE`,
+`WAIT_RULE_FAULT`, `UNMATCHED_FACTS`, `HOST_RETRIES_EXHAUSTED` — means the
+treatment or its evidence chain was not delivered as assigned. Those rows
+still cut the whole task, the design still allows zero primary task cuts, and
+any cut makes H6-timing `NOT_ESTIMABLE`. None of these occurred in the first
+run's 1,890 rows.
+
+## Power and pilot evidence
+
+The binding pilot is the version 12 bound pilot run, transferred by artifact
+hash rather than rerun. The decision rule pins:
+
+- pilot run ID `h6-a30b5cb7dc9174e31329195d`
+- manifest artifact hash `5bce2c0c20fadf77706ae5dbb503d4f938d40e2d4178c5395b36683cfef75e21`
+- power artifact hash `cfaec22f374b353b536dec6067243a1731804f8f1d8d2f4742e1631cf1b27bd5`
+- the pilot's decision-rule hash, preregistration hash, harness source hash,
+  and analysis version as recorded in its immutable manifest
+- simulated H6-timing power **0.8364** at 18 main tasks, against the required
+  0.80
+
+Before any rerun row executes, the runner re-verifies the pilot bundle: the
+manifest artifact hashes must match, the pilot's own frozen version 12 rule
+must hash to the pinned value, and the pilot's statistics and power artifacts
+must replay byte-identically from its immutable rows under the current
+harness. The replay proves the harness changes that added this mode did not
+alter the frozen analysis. The gate is timing power alone; this design has no
+content or no-trap rows to power.
+
+The design quantities that drove the pilot simulation — task count, variants,
+seeds per task, arms, caps, model profile — are identical to the version 12
+main design on the timing pair, so the transferred power estimate applies to
+this run as registered.
+
+## Execution requirements
+
+1. The trap audit must be re-run for the registered model profile under the
+   current harness source hash and decision rule version 13 hash, and must
+   pass the unchanged thresholds, before any main row runs.
+2. The manifest freezes the dataset, split, expected row keys, run order,
+   model profile, seeds, arm configs, note template, caps, statistics seed,
+   and this rule, exactly as version 12 froze them.
+3. A deterministic fake-agent smoke run repeated twice must yield identical
+   row, trace, and decision hashes before live rows run.
+4. Any operational deviation goes in the append-only deviation file. A change
+   to an arm, threshold, task, model profile, seed set, cap, or analysis after
+   the first main row makes the run `NOT_ESTIMABLE` for this manifest.
+5. Rows from this manifest are never pooled with version 12 rows in any
+   confirmatory claim. The first run's estimates remain on its own record.
+
+## Run artifacts
+
+The run writes the same machine-readable artifacts as version 12: `run.json`,
+`expected-design.json`, `checkpoints/`, `episodes.jsonl`, `fact-pair-audit.json`,
+`power.json`, `audit.json`, `deviations.jsonl`, `statistics.json`, and the
+frozen copy of `decision-rule-timing.json`. The raw artifacts, not prose,
+control the decision.

package/dist/index.d.ts

@@ -6752,6 +6752,8 @@ declare const H6_FROZEN_SPLITS: Readonly<{
     pilot: readonly string[];
     main: readonly string[];
 }>;
+declare const H6_FROZEN_EVIDENCE_PATHS: readonly ["decision-rule-timing.json", "trap-audit.json", "tasks/task_01_cache_invalidation", "tasks/task_02_async_queue", "tasks/task_03_token_bucket", "tasks/task_04_event_emitter", "tasks/task_05_trie_search", "tasks/task_06_connection_pool"];
+declare function isFrozenEvidencePath(fixturePath: string): boolean;
 declare const TrapFingerprintV1Schema: z.ZodObject<{
     version: z.ZodLiteral<1>;
     trapId: z.ZodEnum<["flaky-looking-test", "misleading-error-message", "wrong-layer-fix", "hidden-invariant", "stale-cache-illusion", "config-shadowing"]>;
@@ -10474,4 +10476,4 @@ declare function pickOne<T>(rng: SeededRandom, items: readonly T[]): T;
 /** Deterministic Fisher-Yates shuffle returning a new array. */
 declare function shuffled<T>(rng: SeededRandom, items: readonly T[]): T[];
 
-export { AMA_BENCH_DIAGNOSTIC_VARIANTS, ASSISTANT_AGENT_CONFIG_KEY, ASSISTANT_JUDGE_CONFIG_KEY, ASSISTANT_MEETING_PREP_SCENARIOS, ASSISTANT_MEETING_PREP_SMOKE_SCENARIOS, ASSISTANT_MORNING_BRIEF_SCENARIOS, ASSISTANT_MORNING_BRIEF_SMOKE_SCENARIOS, ASSISTANT_NEXT_BEST_ACTION_SCENARIOS, ASSISTANT_NEXT_BEST_ACTION_SMOKE_SCENARIOS, ASSISTANT_RUBRIC_DIMENSIONS, ASSISTANT_RUBRIC_ID_KEY, ASSISTANT_SEEDS_CONFIG_KEY, ASSISTANT_SPOT_CHECK_DIR_KEY, ASSISTANT_SYNTHESIS_SCENARIOS, ASSISTANT_SYNTHESIS_SMOKE_SCENARIOS, type AblationConfigOverrides, type AbstentionRetrievalCase, type ActionIntentV1, ActionIntentV1Schema, type AggregateMetrics, type AmaBenchDiagnosticAdapterOptions, type AmaBenchDiagnosticAnswererMode, type AmaBenchDiagnosticBreakdown, type AmaBenchDiagnosticMatrixArtifact, type AmaBenchDiagnosticRecallMode, type AmaBenchDiagnosticRunContext, type AmaBenchDiagnosticTaskEvidence, type AmaBenchDiagnosticTaskRow, type AmaBenchDiagnosticVariant, type AmaBenchDiagnosticVariantSummary, type AnalyzeRepeatedFailureOptions, type AnthropicProviderConfig, type AssistantAgent, type AssistantMemoryFact, type AssistantMemoryGraph, type AssistantRubricDimension, type AssistantRubricRequest, type AssistantRubricScores, type AssistantRunnerOptions, type AssistantScenario, type AssistantStance, type AttackRecallOptions, type AttackRetrievalHit, type AttackerMode, type AttributeOptions, type AttributionClass, type AttributionEnvironment, type AttributionLabel, type AttributionMemory, type AttributionReport, BENCHMARK_ARTIFACT_SCHEMA_VERSION, BENCHMARK_INTEGRITY_META_SCHEMA, BENCHMARK_REPRO_MANIFEST_FILENAME, BENCHMARK_REPRO_MANIFEST_SCHEMA_VERSION, BENCHMARK_RESULT_SCHEMA, BENCHMARK_SPLIT_TYPES, BUILD_WEEK_EVIDENCE_RECEIPT_SCHEMA_VERSION, BUILD_WEEK_LIMITATIONS, type BaseTask, BaseTaskSchema, type BaselineRow, type BaselineScenario, type BeamDatasetPreview, type BenchAggregateMetric, type BenchAssistantTaskDetails, type BenchConfig, type BenchIntegritySplit, type BenchIntegritySummary, type BenchJudge, type BenchJudgeResult, type BenchMemoryAdapter, type BenchMetricHighlight, type BenchModelSource, type BenchPerSeedScore, type BenchReasoningEffort, type BenchRecallLineageStatus, type BenchRecallOptions, type BenchRecallSupportAssessment, type BenchRecallSupportRequest, type BenchRecallSupportStatus, type BenchRecallTrace, type BenchRecallTraceCoreCapture, type BenchRecallTraceLcmCandidate, type BenchRecallTraceRange, type BenchRecallTraceSection, type BenchRecallTraceSelection, type BenchRecallWithTraceResult, type BenchResponder, type BenchResponse, type BenchResultFileWarning, type BenchResultSummary, type BenchResultSummaryPayload, type BenchRuntimeProfile, type BenchTaskScoreEntry, type BenchTaskSummary, type BenchTier, type BenchmarkArtifact, type BenchmarkArtifactEnvironment, type BenchmarkArtifactHardware, type BenchmarkArtifactJudgeCalibration, type BenchmarkArtifactPerTaskScore, type BenchmarkArtifactSystem, type BenchmarkArtifactTier, type BenchmarkCategory, type BenchmarkDefinition, type BenchmarkExecutionProvenance, type BenchmarkIntegrityMeta, type BenchmarkMeta, type BenchmarkMode, type BenchmarkReport, type BenchmarkReproManifest, type BenchmarkReproManifestDataset, type BenchmarkReproManifestFile, type BenchmarkReproManifestResult, type BenchmarkReproManifestSupplementalArtifact, type BenchmarkResult, type BenchmarkSplitType, type BenchmarkStatus, type BenchmarkSuiteResult, type BenchmarkTier, type BootstrapKappaOptions, type BootstrapKappaResult, type BuildBenchmarkArtifactInput, type BuildBenchmarkPublishFeedOptions, type BuildBenchmarkReproManifestOptions, type BuildBuildWeekEvidenceReceiptOptions, type BuildWeekEvidenceReceipt, type BuildWeekEvidenceReceiptProvider, type BuildWeekLimitationCode, type BuiltInProvider, CALIBRATION_SLICE_SIZE, CANARY_FIXED_RECALL, CANARY_SCORE_FLOOR, DEFAULT_10K_FIXTURE as CODING_GRAPH_10K_FIXTURE, CODING_GRAPH_BENCH_SCHEMA_VERSION, DEFAULT_TOLERANCE_PERCENT as CODING_GRAPH_DEFAULT_TOLERANCE, MIN_ITERATIONS as CODING_GRAPH_MIN_ITERATIONS, DEFAULT_SMOKE_FIXTURE as CODING_GRAPH_SMOKE_FIXTURE, type CalibrationAnswer, type CalibrationVerdictPair, type CanaryAdapterOptions, type CanaryFloorCheck, type CaptureLoCoMoRetrievalTraceOptions, type ClaudeCliProviderConfig, type CodexCliNativeUsage, type CodexCliProviderConfig, CodexCreditAccountingError, type CodexCreditBudgetConfig, CodexCreditDispatchError, type CodexCreditReceipt, type CodexCreditReceiptScope, type CodexCreditReconciliationReceipt, type CodingGraphBaseline, type CodingGraphBenchConfig, type CodingGraphBenchReport, type MachineFingerprint as CodingGraphMachineFingerprint, type CodingGraphMetricKey, type RegressionMetricDetail as CodingGraphRegressionDetail, type RegressionMetricKey as CodingGraphRegressionKey, type RegressionGateResult as CodingGraphRegressionResult, type CohenKappaResult, type ComparisonMetricDelta, type ComparisonResult, type CompletionOpts, type CompletionResult, type ConfidenceInterval, type ContaminationCheckResult, type ContaminationEntry, type ContaminationManifest, type ControlledGateDecision, type ControlledResponsesAgentDriverConfig, type ControlledResponsesCaps, type ControlledResponsesDisposition, ControlledResponsesDriver, type ControlledResponsesDriverConfig, type ControlledResponsesEpisodeInput, type ControlledResponsesEpisodeResult, type ControlledResponsesFault, type ControlledResponsesResponseEvent, type ControlledResponsesToolDefinition, type ControlledResponsesToolEvent, type ControlledResponsesTransport, type CustomBenchmarkScoring, type CustomBenchmarkSpec, type CustomBenchmarkTask, DATASET_SPLITS, DEFAULT_ABLATION_BENCHMARK, DEFAULT_ABLATION_BOOTSTRAP_SEED, DEFAULT_ASSISTANT_RUBRIC_ID, DEFAULT_ATTACKER_PROMPT_PATH_GETTER, DEFAULT_BASELINE_SCENARIOS, DEFAULT_JUDGE_BINARIZATION_THRESHOLD, DEFAULT_KAPPA_BOOTSTRAP_SAMPLES, DEFAULT_KAPPA_CONFIDENCE_LEVEL, DEFAULT_OPENAI_RESPONSES_JUDGE_MODEL, DRIFT_GEN_DEFAULTS, DRIFT_GEN_VERSION, type DatasetSource, type DatasetSplit, type DiagnoseLoComoProfileDeltaOptions, type DiagnoseLoComoRecallDeltaOptions, type DiscoveredModel, type DriftGenAuditRecord, type DriftGenCorpus, type DriftGenManifest, type DriftGenOptions, type DriftGenResult, type DriftSession, type DriftSessionTurn, type DriftValidationReport, type DriftValidationStats, EMPTY_CONTAMINATION_MANIFEST, type EffectSizeInterpretation, type EffectSizeSummary, type EvaluateTaskStateOptions, type ExplainResult, type ExtractedEntity, type ExtractedLink, type ExtractedPage, type ExtractionAttackOptions, type ExtractionAttackResult, type ExtractionAttackTarget, type FixtureGenerator, type FixtureOutput, type FixtureVariant, GATE_STATUSES, GENERAL_ANSWER_JUDGE_RUBRIC, type GateStatus, type GeneratedFile, type GeneratedRepo, type GoldEntity, type GoldEntityType, type GoldFact, type GoldFactKind, type GoldGraph, type GoldLink, type GoldMemoryAttribution, type GoldPage, type GoldProbe, type GoldProbeCategory, H5_PUBLICATION_ANALYSIS_RULE, H5_PUBLICATION_DEFENSE_BASELINES, type H6BenchmarkDataset, H6BenchmarkDatasetSchema, type H6TrapId, H6_ACTION_INTENT_JSON_SCHEMA, H6_ARMS, H6_DATASET_JSON_SCHEMA, H6_DECISION_RULE, H6_FROZEN_INVENTORY_HASH, H6_FROZEN_SEED, H6_FROZEN_SPLITS, H6_SUPPORT_ARTIFACT_PATHS, H6_TASK_JSON_SCHEMA, H6_TRAP_FINGERPRINT_JSON_SCHEMA, H6_TRAP_IDS, HOST_FAULT_RETRY_LIMIT, type HarnessRng, INJECTION_SUITE_ARMS, INJECTION_SUITE_FAMILIES, INJECTION_SUITE_PUBLICATION_ARMS, INJECTION_SUITE_STAGES, INJECTION_SUITE_VERSION, INTEGRITY_CIPHER_ALGORITHM, INTEGRITY_HASH_ALGORITHM, INTEGRITY_META_FIELDS, INVENTED_DOMAINS, type IngestionBenchAdapter, type IngestionLog, type InjectionSuiteArm, type InjectionSuiteCliInput, type InjectionSuiteCliResult, type InjectionSuiteEpisodeRow, type InjectionSuiteFamily, type InjectionSuiteOnlineAdaptiveInput, type InjectionSuiteProductEvidence, type InjectionSuiteRowIdentity, type InjectionSuiteStage, JUDGE_CALIBRATION_KAPPA_THRESHOLD, JUDGE_CALIBRATION_PROTOCOL_VERSION, type JudgeCalibrationCheckpointProvenance, type JudgeCalibrationIdentities, type JudgeCalibrationResult, type JudgeCategory, type KappaConfidenceInterval, LOCAL_LAB_PROVIDER_KINDS, LOCOMO_DATASET_FILENAMES, LOCOMO_FULL_TASK_COUNT, LOCOMO_RECALL_DIFF_LINE_LIMIT, LOCOMO_RECALL_EXCERPT_CHARS, LOCOMO_RETRIEVAL_TRACE_BUDGET_VERSION, LOCOMO_RETRIEVAL_TRACE_DELTA_SCHEMA_VERSION, LOCOMO_RETRIEVAL_TRACE_SCHEMA_VERSION, LOCOMO_RETRIEVAL_TRACE_SELECTION_VERSION, LONG_MEM_EVAL_DATASET_FILENAMES, type LeaderboardArtifactWrite, type LettaAdapterConfig, LettaMemCorrectAdapter, type LlmJudge, type LlmProvider, type LoCoMoCategory, type LoCoMoRetrievalMechanism, type LoCoMoRetrievalMechanismSummary, type LoCoMoRetrievalSessionReceipt, type LoCoMoRetrievalStructuralTrace, type LoCoMoRetrievalTaskDelta, type LoCoMoRetrievalTaskReceipt, type LoCoMoRetrievalTraceCoreCaptureReceipt, type LoCoMoRetrievalTraceDeltaReport, type LoCoMoRetrievalTraceProfile, type LoCoMoRetrievalTraceReceipt, type LoCoMoRetrievalTraceSelectionManifest, type LoCoMoRetrievalTraceSelector, type LoCoMoStructuralMultisetDelta, type LoComoCategoryDelta, type LoComoFinalContextRegression, type LoComoMetricDelta, type LoComoProfileArtifactEvidence, type LoComoProfileDeltaReport, type LoComoRawResultEvidence, type LoComoRecallCategoryDelta, type LoComoRecallContextSummary, type LoComoRecallDeltaReport, type LoComoRecallLineDelta, type LoComoRecallLineEvidence, type LoComoRecallMetricDelta, type LoComoRecallResultProvenance, type LoComoRecallTextDigest, type LoComoTaskRegression, type LoadDatasetOptions, type LoadSealedQrelsOptions, type LoadedDataset, type LoadedJudgeCalibrationState, type LocalLabManifest, type LocalLabManifestNotes, type LocalLabPhase, type LocalLabPhaseDescriptor, type LocalLabPhaseExecute, type LocalLabPhaseName, type LocalLabPhaseOutcome, LocalLabPreflightError, type LocalLabPreflightFailure, type LocalLabPreflightInput, type LocalLabPreflightOptions, type LocalLabPreflightResult, type LocalLabPreflightSuccess, type LocalLabProviderKind, type LocalLabRoleConfig, type LocalLlmProviderConfig, MAX_ROW_ATTEMPTS, MEMCORRECT_CORRECTION_ACCEPTANCE_RUBRIC, MEMCORRECT_CORRECTION_ACCEPTANCE_RUBRIC_VERSION, MEMCORRECT_STALE_HARM_RUBRIC, MEMCORRECT_STALE_HARM_RUBRIC_VERSION, MEMORY_EVAL_DIMENSIONS, MEMORY_EVAL_PUBLIC_LINE, MIN_CALIBRATION_SOURCE_TASKS, MITIGATED_BASELINE_SCENARIOS, type MaterializeOptions, type MaterializedRepo, type McpArgumentSemantic, type McpBackendErrorCode, type McpBackendResult, type McpBenchMemoryAdapter, type McpConformanceResult, type McpHttpTransportConfig, type McpListedTool, type McpMemCorrectAdapter, type McpMemoryAdapterOptions, McpMemoryBackendError, type McpMemoryToolMapping, type McpMemoryTransportConfig, type McpStdioTransportConfig, type McpToolCallResult, type McpToolClient, type McpToolMappingEntry, type McpToolMappingValue, type McpToolOperation, type Mem0AdapterConfig, Mem0MemCorrectAdapter, type MemCorrectGeneratorOptions, type MemCorrectJudgeRequest, type MemCorrectJudgeResult, type MemCorrectSystemAdapter, type MemoryEvalCategory, type MemoryEvalDimension, type MemoryEvalDimensionId, type MemoryEvalMetric, type MemoryGraph, type MemoryStats, type MemorySystem, type Message, type MetricAggregate, type MicroMetric, MissingCredentialError, type MitigatedBaselineConfig, type MitigatedTargetConfig, type MultipleChoiceQuestion, ONLINE_ADAPTIVE_STAGE, OPENAI_RESPONSES_JUDGE_RUBRIC_VERSION, OTHER_NAMESPACE_MEMORIES, type OllamaChatMessage, type OllamaProviderConfig, type OnlineAdaptiveCorpusLine, type OnlineAdaptiveCorpusManifest, type OnlineAdaptiveStatistics, type OpenAiCompatibleProviderConfig, OpenAiResponsesJudgeError, type OpenAiResponsesJudgeErrorCode, type OpenAiResponsesJudgeTelemetry, OpenAiResponsesProvider, type OpenAiResponsesProviderConfig, type OpenAiResponsesVerdict, type OpenAiResponsesVerdictResult, PROCEDURAL_REAL_SCENARIOS, PROCEDURAL_REAL_SCENARIOS_SMOKE, PUBLISHED_BENCHMARK_ARTIFACT_IDS, type PairedAnswerReplayCache, type PairedAnswerReplayEntry, type ParsedOllamaChatResponse, type PersonalizationRetrievalCase, type PreflightDiscoveredModel, type ProceduralAblationArtifact, type ProceduralAblationPerCase, type ProceduralAblationScenario, type ProceduralRealScenario, type ProceduralRealScenarioCategory, type ProviderBaseConfig, type ProviderConfig, type ProviderDiscoveryResult, type ProviderFactoryConfig, type PublishSkipReason, type PublishSkipRecord, type PublishedBenchmarkFeed, type PublishedBenchmarkFeedEntry, type PublishedBenchmarkId, REPEATED_FAILURE_ARMS, REPEATED_FAILURE_CONFIDENCE_LEVEL, REPEATED_FAILURE_INVALID_REASONS, REPEATED_FAILURE_STATISTICS_DRAWS, REQUIRED_FRONTMATTER_FIELDS, type RecallMetrics, type RecoveredMemory, type RegressionDetail, type RegressionGateResult$1 as RegressionGateResult, type RemnicAdapterOptions, type RepeatedFailureActionEvaluator, type RepeatedFailureArm, type RepeatedFailureCheckpointLoadResult, type RepeatedFailureCliCommandResult, type RepeatedFailureEffectAnalysis, type RepeatedFailureEpisode, type RepeatedFailureEpisodeDriver, type RepeatedFailureEpisodeEvidence, type RepeatedFailureEpisodeInput, type RepeatedFailureEpisodeRow, type RepeatedFailureExpectedDesign, type RepeatedFailureFactPairAudit, type RepeatedFailureFinalRepoEvidence, type RepeatedFailureFinalState, type RepeatedFailureGateEvent, type RepeatedFailureHolmResult, type RepeatedFailureInterval, type RepeatedFailureInvalidReason, type RepeatedFailureIsolationIdentity, type RepeatedFailureLocalToolHost, type RepeatedFailureNullableInterval, RepeatedFailureOllamaChatDriver, type RepeatedFailureOllamaChatDriverConfig, type RepeatedFailureProposedAction, type RepeatedFailureRowCheckpoint, type RepeatedFailureRowClaim, type RepeatedFailureRowIdentity, RepeatedFailureRowStore, type RepeatedFailureRowStoreOptions, type RepeatedFailureRunMetadata, type RepeatedFailureStatisticalAnalysis, type RepeatedFailureSuiteManifest, type RepeatedFailureSupportDecision, type RepeatedFailureTaskCut, type RepeatedFailureTimidityAnalysis, type RepeatedFailureTokenUsage, type RepeatedFailureTokenizer, type RepeatedFailureToolDefinition, type RepeatedFailureToolExecutionResult, type RepeatedFailureTrapAuditArtifact, type RepeatedFailureTrapAuditExpected, type RepeatedFailureTrapAuditMetrics, type RepeatedFailureTrapAuditRow, type RepeatedFailureTrapAuditRowIdentity, type RepeatedFailureTrapAuditThresholds, type RepeatedFailureTry, type ReplayRepeatedFailureStatisticsOptions, type ReportCardProvenanceContext, type ResolveBenchRuntimeProfileOptions, type ResolvedBenchRuntimeProfile, type ResolvedLocalLabProfile, type ResolvedLocalLabRole, type ResolvedRunBenchmarkOptions, type ResponsesApiOutputItem, type ResponsesApiRequest, type ResponsesApiResponse, type ResponsesApiUsage, type RetrievalMissStage, type RevisionShas, type RotatedChoices, type RunBenchmarkOptions, type RunJudgeCalibrationOptions, type RunProceduralAblationCliArgs, type RunProceduralAblationOptions, type RunRepeatedFailureCliCommandInput, type RunRepeatedFailureSuiteOptions, type RunRepeatedFailureSuiteResult, type RunSequentialPhasesOptions, type RunTrapAuditOptions, SCHEMA_TIER_FIXTURE, SCHEMA_TIER_SMOKE_FIXTURE, SEALED_PROMPT_REGISTRY, SINGLE_FLAG_ABLATION_MATRIX, STATE_CLASSIFICATIONS, SYNTHETIC_MEMORIES, type SanitizedDiagnosticProvider, type SavedBaseline, type SchemaTierCorpus, type SchemaTierFixture, type SchemaTierName, type SchemaTierPage, type SchemaTierPageFrontmatter, type SealedArtifact, type SealedJudgeDecision, type SealedJudgeInput, type SealedQrelsArtifact, type SealedQrelsHandle, type SealedRubric, type SearchResult, type SeededMemory, type SeededRandom, type SeededRng, type SequentialPhaseHooks, type SingleFlagAblationCell, type SingleFlagAblationId, type SpotCheckLogger, type StageObservation, type StageStatus, type StateClassification, type StateEvaluationResult, type StatisticalReport, type StrategyPatch, StrategyPatchSchema, type StructuredJudge, StructuredJudgeError, type StructuredJudgeErrorCode, type StructuredJudgeProvider, type StructuredJudgeTelemetry, type StructuredJudgeVerdict, type StructuredJudgeVerdictResult, type StructuredVerdictRequest, type SyntheticEdge, type SyntheticEmailIngestionAdapterOptions, type SyntheticFile, type SyntheticFileIR, SyntheticFileSchema, type SyntheticRepoConfig, type SyntheticSymbol, type SyntheticTargetOptions, TRAP_TAXONOMY, type TaskAttribution, type TaskAttributionGoldWitnessV1, type TaskAttributionRetrievalWitnessV1, type TaskAttributionWitness, type TaskAttributionWitnessRuntimeV1, type TaskAttributionWitnessV1, type TaskResult, type TaskTokenUsage, type TaskVariant, TaskVariantSchema, type TemporalRetrievalCase, type ThirdPartyAdapterConfig, type TierDetail, type TimelineEntry, type TokenUsage, type TrapFingerprintV1, TrapFingerprintV1Schema, type TrapTaxonomyItem, TrapTaxonomyItemSchema, type ValidationIssue, type ValidationReport, type WallMetric, type WriteBenchmarkArtifactResult, type ZepAdapterConfig, ZepMemCorrectAdapter, addContaminationEntry, aggregateTaskScores, analyzeInjectionSuiteOnlineAdaptiveRun, analyzeInjectionSuitePublicationRows, analyzeInjectionSuitePublicationRun, analyzeInjectionSuitePublicationUtility, analyzeInjectionSuitePublicationUtilityFile, analyzeInjectionSuiteRun, analyzeInjectionSuiteUtility, analyzeRepeatedFailureRows, answerBenchmarkQuestion, applyPatchAndCommit, assertCanaryUnderFloor, assertIntegrityMetaPresent, assertPublishableIntegrity, assertSha256Hex, assertTrapDatasetPreflight, assistantMeetingPrepDefinition, assistantMorningBriefDefinition, assistantNextBestActionDefinition, assistantSynthesisDefinition, attributeGoldMemory, attributeRun, attributeTask, backlinkF1, binarizeJudgeScore, bootstrapCohensKappaConfidenceInterval, bootstrapMeanConfidenceInterval, buildAmaBenchDiagnosticMatrixArtifact, buildAmaBenchDiagnosticVariantSummary, buildAmaBenchLeaderboardRows, buildBaselineFromReport, buildBenchmarkArtifact, buildBenchmarkArtifactFilename, buildBenchmarkPublishFeed, buildBenchmarkReproManifest, buildBenchmarkRunSeeds, buildBuildWeekEvidenceReceipt, buildCodexCreditReceipt, buildDriftCorpus, buildJudgePayload, buildOracleTrajectoryRecall, buildProviderFreeLoCoMoRetrievalConfig, buildRepeatedFailureRowKey, buildSchemaTierFixture, buildSchemaTierSmokeFixture, calculateCodexBudgetUnits, calculateJaccardSimilarity, calendarFixture, canonicalJsonStringify, captureBenchmarkExecutionProvenance, captureLoCoMoRetrievalTrace, captureMachineFingerprint, chatFixture, checkCodingGraphRegression, checkDatasetContamination, checkRegression, clampScore, cohensD, compareResults, computeBenchmarkReproDatasetInventoryHash, computeBenchmarkReproManifestArtifactHash, computeCohensKappa, computeH6InventoryHash, computeH6SupportArtifactHashes, computeRevisionShas, computeSealHash, computeTrapAuditArtifactHash, computeTrapAuditMetrics, containsAnswer, createSeededRng$1 as createAdamSeededRng, createAmaBenchDiagnosticAdapter, createAnthropicProvider, createCanaryAdapter, createClaudeCliProvider, createCodexCliProvider, createSeededRng as createCodingGraphSeededRng, createControlledResponsesAgentDriver, createDeterministicSpotCheckLogger, createGatewayResponder, createLightweightAdapter, createLiteLlmProvider, createLocalLlmProvider, createMcpDemoMemCorrectAdapter, createMcpDemoMemoryAdapter, createMcpMemCorrectAdapter, createMcpMemoryAdapter, createMitigatedTarget, createOllamaProvider, createOpenAiCompatibleProvider, createOpenAiResponsesBenchJudge, createOpenAiResponsesProvider, createSeededRandom$1 as createProceduralAblationSeededRandom, createProvider, createProviderBackedAmaBenchRecommendedJudge, createProviderBackedJudge, createProviderBackedResponder, createProviderBackedStructuredJudge, createRemnicAdapter, createRepeatedFailureOllamaChatDriver, createResponderFromProvider, createSeededRandom, createSeededRng$2 as createSeededRng, createSpotCheckFileLogger, createStructuredBenchJudge, createStructuredJudgeFromProvider, createSyntheticEmailIngestionAdapter, createSyntheticTarget, createTimeoutGuardedAdapter, decideInjectionSuiteCampaign, decideRepeatedFailureContent, decideRepeatedFailureStudy, decideRepeatedFailureTiming, defaultBenchmarkBaselineDir, defaultBenchmarkPublishPath, deleteBenchmarkResults, diagnoseLoCoMoRetrievalTraceDelta, diagnoseLoComoProfileDelta, diagnoseLoComoRecallDelta, discoverAllProviders, discoveryEndpointFor, emailFixture, entityRecall, evaluateTaskState, exactMatch, executeLocalRow, extractMetrics as extractCodingGraphMetrics, extractContentWords, extractMarkdownSectionsByTitle, f1Score, fixtureToAblationScenarios, formatHandoffNote, formatMissingDatasetError, generateDriftCorpus, generateFamilyVariants, generateH6BenchmarkDataset, generateReport, generateSuiteVariants, generateSyntheticRepo, getAblationCell, getBenchmark, getBenchmarkLowerIsBetter, getGitSha, getMemoryEvalDimension, getProviderBackedJudgePromptIdentity, getRemnicVersion, getTrapTaxonomyItem, hashBenchmarkArtifact, hashBytes, hashCanonicalJson, hashOrderedQuestionIds, hashString, holmAdjust, injectionSuiteResumeContractHash, integrityMetaIsComplete, interpretEffectSize, isAmaBenchUnknownLikeAnswer, isContaminationEntry, isContaminationManifest, isRepeatedFailureTimidityEquivalent, isSafeSyntheticPath, isSealedQrelsArtifact, isSha256Hex, isStructuredJudgeProvider, isTaskFailed, judgeMemCorrectCorrectionAcceptance, judgeMemCorrectStaleMemoryHarm, lexicalSimilarity, linkMatches, listBenchmarkBaselines, listBenchmarkResults, listBenchmarks, listMemoryEvalBenchmarkIds, listMemoryEvalDimensions, llmJudgeScore, llmJudgeScoreDetailed, loadAblationFixture, loadBaseline, loadBeamDatasetPreview, loadBenchmarkArtifact, loadBenchmarkBaseline, loadBenchmarkReportCardProvenance, loadBenchmarkResult, loadBenchmarkResultSummaries, loadCommittedH6BenchmarkDataset, loadCustomBenchmarkFile, loadJudgeCalibrationState, loadLoCoMo10, loadLocalLabManifest, loadLongMemEvalS, loadSealKeyFromEnv, loadSealedQrels, loadSealedRubric, matchEntity, materializeTaskRepo, mergeContaminationManifests, openSeal, orchestrateBenchmarkRuns, pairedDeltaConfidenceInterval, parseBenchmarkArtifact, parseCanaryFloor, parseCodexJsonlUsage, parseCustomBenchmark, parseLocalLabManifest, parseRepeatedFailureEpisodeRow, parseRubricResponse, parseSealedQrels, pickOne, pickStableQualifiedName, planInjectionSuiteRows, planOnlineAdaptiveRows, precisionAtK, preflightLoCoMoRetrievalTraceCapture, preflightLocalLabRole, projectFolderFixture, randomInt, recallAtK, reconcileCodexCreditLedger, redactBenchmarkResultSecrets, relativeRiskReduction, renderAttributionReportTable, renderBaselineMarkdown, renderBenchmarkResultExport, renderLoComoProfileDeltaMarkdown, renderLoComoRecallDeltaMarkdown, renderMemorySummaryForJudge, renderMemoryViewForAgent, replayInjectionSuiteStatistics, replayRepeatedFailureStatistics, resolveAssistantAgent, resolveAssistantRubricId, resolveAssistantSeeds, resolveAssistantSpotCheckDir, resolveBenchRuntimeProfile, resolveBenchmarkPhaseTimeoutMs, resolveBenchmarkProgressLogging, resolveBenchmarkResultReference, resolveBenchmarkRunCount, resolveBenchmarkRunId, resolveCanaryFloorFromEnv, resolveCodexCreditBudgetConfig, resolveCommittedH6FixtureDirectory, resolveEffectiveCanaryFloor, resolveLocalLabJudgeProviderConfig, resolveLocalLabProfile, resolveLocalLabRole, resolveStructuredJudge, rotateDistractors, rougeL, runAssistantBenchmark, runAssistantMeetingPrepBenchmark, runAssistantMorningBriefBenchmark, runAssistantNextBestActionBenchmark, runAssistantSynthesisBenchmark, runAttributeCliCommand, runBaseline, runBenchSuite, runBenchmark, runCodingGraphBenchmark, runCustomBenchmarkFile, runDriftGenCliCommand, runExplain, runExtractionAttack, runInjectionSuiteCliCommand, runInjectionSuiteOnlineAdaptive, runInjectionSuiteUtility, runJudgeCalibration, runMitigatedBaseline, runProceduralAblation, runProceduralAblationCli, runRepeatedFailureCliCommand, runRepeatedFailurePaperReportCliCommand, runRepeatedFailureSuite, runSealedJudge, runSequentialPhases, runTrapAudit, runTrapAuditCliCommand, runWithinCodexCreditBudget, safeHexEqual, sanitizeBenchmarkResultForJson, sanitizeLoComoResultReference, saveBaseline, saveBenchmarkBaseline, schemaCompleteness, sealPayload, selectAmaBenchDiagnosticVariants, selectCalibrationSlice, selectFixtureVariant, serializeAttributionReport, serializeBenchmarkArtifact, serializeBuildWeekEvidenceReceipt, serializeH6FixtureJson, serializeJsonl, serializeLoCoMoRetrievalTraceDelta, serializeLoCoMoRetrievalTraceReceipt, serializeSealedQrels, shuffleTasks, shuffled, stripCodeFences, summarizeBenchmarkResult, timed, tokenizeContent, unresolvedHelperImports, validateDriftCorpus, validateH6Dataset, validateH6FixtureBundle, validateH6StateDefiningIndependence, validateOllamaChatEndpoint, verifyMatchingTrapAudit, verifyRubricDigest, verifyTrapAuditArtifact, writeBenchmarkArtifact, writeBenchmarkPublishFeed, writeBenchmarkReproManifest, writeBenchmarkResult, writeBuildWeekEvidenceReceipt, writeH6FixtureBundle, writeJudgeCalibrationState, writeLeaderboardArtifactsForResult, writeRepeatedFailurePaperArtifacts, writeRepeatedFailureRunMetadata, writeRepeatedFailureStatistics, zeroScores };
+export { AMA_BENCH_DIAGNOSTIC_VARIANTS, ASSISTANT_AGENT_CONFIG_KEY, ASSISTANT_JUDGE_CONFIG_KEY, ASSISTANT_MEETING_PREP_SCENARIOS, ASSISTANT_MEETING_PREP_SMOKE_SCENARIOS, ASSISTANT_MORNING_BRIEF_SCENARIOS, ASSISTANT_MORNING_BRIEF_SMOKE_SCENARIOS, ASSISTANT_NEXT_BEST_ACTION_SCENARIOS, ASSISTANT_NEXT_BEST_ACTION_SMOKE_SCENARIOS, ASSISTANT_RUBRIC_DIMENSIONS, ASSISTANT_RUBRIC_ID_KEY, ASSISTANT_SEEDS_CONFIG_KEY, ASSISTANT_SPOT_CHECK_DIR_KEY, ASSISTANT_SYNTHESIS_SCENARIOS, ASSISTANT_SYNTHESIS_SMOKE_SCENARIOS, type AblationConfigOverrides, type AbstentionRetrievalCase, type ActionIntentV1, ActionIntentV1Schema, type AggregateMetrics, type AmaBenchDiagnosticAdapterOptions, type AmaBenchDiagnosticAnswererMode, type AmaBenchDiagnosticBreakdown, type AmaBenchDiagnosticMatrixArtifact, type AmaBenchDiagnosticRecallMode, type AmaBenchDiagnosticRunContext, type AmaBenchDiagnosticTaskEvidence, type AmaBenchDiagnosticTaskRow, type AmaBenchDiagnosticVariant, type AmaBenchDiagnosticVariantSummary, type AnalyzeRepeatedFailureOptions, type AnthropicProviderConfig, type AssistantAgent, type AssistantMemoryFact, type AssistantMemoryGraph, type AssistantRubricDimension, type AssistantRubricRequest, type AssistantRubricScores, type AssistantRunnerOptions, type AssistantScenario, type AssistantStance, type AttackRecallOptions, type AttackRetrievalHit, type AttackerMode, type AttributeOptions, type AttributionClass, type AttributionEnvironment, type AttributionLabel, type AttributionMemory, type AttributionReport, BENCHMARK_ARTIFACT_SCHEMA_VERSION, BENCHMARK_INTEGRITY_META_SCHEMA, BENCHMARK_REPRO_MANIFEST_FILENAME, BENCHMARK_REPRO_MANIFEST_SCHEMA_VERSION, BENCHMARK_RESULT_SCHEMA, BENCHMARK_SPLIT_TYPES, BUILD_WEEK_EVIDENCE_RECEIPT_SCHEMA_VERSION, BUILD_WEEK_LIMITATIONS, type BaseTask, BaseTaskSchema, type BaselineRow, type BaselineScenario, type BeamDatasetPreview, type BenchAggregateMetric, type BenchAssistantTaskDetails, type BenchConfig, type BenchIntegritySplit, type BenchIntegritySummary, type BenchJudge, type BenchJudgeResult, type BenchMemoryAdapter, type BenchMetricHighlight, type BenchModelSource, type BenchPerSeedScore, type BenchReasoningEffort, type BenchRecallLineageStatus, type BenchRecallOptions, type BenchRecallSupportAssessment, type BenchRecallSupportRequest, type BenchRecallSupportStatus, type BenchRecallTrace, type BenchRecallTraceCoreCapture, type BenchRecallTraceLcmCandidate, type BenchRecallTraceRange, type BenchRecallTraceSection, type BenchRecallTraceSelection, type BenchRecallWithTraceResult, type BenchResponder, type BenchResponse, type BenchResultFileWarning, type BenchResultSummary, type BenchResultSummaryPayload, type BenchRuntimeProfile, type BenchTaskScoreEntry, type BenchTaskSummary, type BenchTier, type BenchmarkArtifact, type BenchmarkArtifactEnvironment, type BenchmarkArtifactHardware, type BenchmarkArtifactJudgeCalibration, type BenchmarkArtifactPerTaskScore, type BenchmarkArtifactSystem, type BenchmarkArtifactTier, type BenchmarkCategory, type BenchmarkDefinition, type BenchmarkExecutionProvenance, type BenchmarkIntegrityMeta, type BenchmarkMeta, type BenchmarkMode, type BenchmarkReport, type BenchmarkReproManifest, type BenchmarkReproManifestDataset, type BenchmarkReproManifestFile, type BenchmarkReproManifestResult, type BenchmarkReproManifestSupplementalArtifact, type BenchmarkResult, type BenchmarkSplitType, type BenchmarkStatus, type BenchmarkSuiteResult, type BenchmarkTier, type BootstrapKappaOptions, type BootstrapKappaResult, type BuildBenchmarkArtifactInput, type BuildBenchmarkPublishFeedOptions, type BuildBenchmarkReproManifestOptions, type BuildBuildWeekEvidenceReceiptOptions, type BuildWeekEvidenceReceipt, type BuildWeekEvidenceReceiptProvider, type BuildWeekLimitationCode, type BuiltInProvider, CALIBRATION_SLICE_SIZE, CANARY_FIXED_RECALL, CANARY_SCORE_FLOOR, DEFAULT_10K_FIXTURE as CODING_GRAPH_10K_FIXTURE, CODING_GRAPH_BENCH_SCHEMA_VERSION, DEFAULT_TOLERANCE_PERCENT as CODING_GRAPH_DEFAULT_TOLERANCE, MIN_ITERATIONS as CODING_GRAPH_MIN_ITERATIONS, DEFAULT_SMOKE_FIXTURE as CODING_GRAPH_SMOKE_FIXTURE, type CalibrationAnswer, type CalibrationVerdictPair, type CanaryAdapterOptions, type CanaryFloorCheck, type CaptureLoCoMoRetrievalTraceOptions, type ClaudeCliProviderConfig, type CodexCliNativeUsage, type CodexCliProviderConfig, CodexCreditAccountingError, type CodexCreditBudgetConfig, CodexCreditDispatchError, type CodexCreditReceipt, type CodexCreditReceiptScope, type CodexCreditReconciliationReceipt, type CodingGraphBaseline, type CodingGraphBenchConfig, type CodingGraphBenchReport, type MachineFingerprint as CodingGraphMachineFingerprint, type CodingGraphMetricKey, type RegressionMetricDetail as CodingGraphRegressionDetail, type RegressionMetricKey as CodingGraphRegressionKey, type RegressionGateResult as CodingGraphRegressionResult, type CohenKappaResult, type ComparisonMetricDelta, type ComparisonResult, type CompletionOpts, type CompletionResult, type ConfidenceInterval, type ContaminationCheckResult, type ContaminationEntry, type ContaminationManifest, type ControlledGateDecision, type ControlledResponsesAgentDriverConfig, type ControlledResponsesCaps, type ControlledResponsesDisposition, ControlledResponsesDriver, type ControlledResponsesDriverConfig, type ControlledResponsesEpisodeInput, type ControlledResponsesEpisodeResult, type ControlledResponsesFault, type ControlledResponsesResponseEvent, type ControlledResponsesToolDefinition, type ControlledResponsesToolEvent, type ControlledResponsesTransport, type CustomBenchmarkScoring, type CustomBenchmarkSpec, type CustomBenchmarkTask, DATASET_SPLITS, DEFAULT_ABLATION_BENCHMARK, DEFAULT_ABLATION_BOOTSTRAP_SEED, DEFAULT_ASSISTANT_RUBRIC_ID, DEFAULT_ATTACKER_PROMPT_PATH_GETTER, DEFAULT_BASELINE_SCENARIOS, DEFAULT_JUDGE_BINARIZATION_THRESHOLD, DEFAULT_KAPPA_BOOTSTRAP_SAMPLES, DEFAULT_KAPPA_CONFIDENCE_LEVEL, DEFAULT_OPENAI_RESPONSES_JUDGE_MODEL, DRIFT_GEN_DEFAULTS, DRIFT_GEN_VERSION, type DatasetSource, type DatasetSplit, type DiagnoseLoComoProfileDeltaOptions, type DiagnoseLoComoRecallDeltaOptions, type DiscoveredModel, type DriftGenAuditRecord, type DriftGenCorpus, type DriftGenManifest, type DriftGenOptions, type DriftGenResult, type DriftSession, type DriftSessionTurn, type DriftValidationReport, type DriftValidationStats, EMPTY_CONTAMINATION_MANIFEST, type EffectSizeInterpretation, type EffectSizeSummary, type EvaluateTaskStateOptions, type ExplainResult, type ExtractedEntity, type ExtractedLink, type ExtractedPage, type ExtractionAttackOptions, type ExtractionAttackResult, type ExtractionAttackTarget, type FixtureGenerator, type FixtureOutput, type FixtureVariant, GATE_STATUSES, GENERAL_ANSWER_JUDGE_RUBRIC, type GateStatus, type GeneratedFile, type GeneratedRepo, type GoldEntity, type GoldEntityType, type GoldFact, type GoldFactKind, type GoldGraph, type GoldLink, type GoldMemoryAttribution, type GoldPage, type GoldProbe, type GoldProbeCategory, H5_PUBLICATION_ANALYSIS_RULE, H5_PUBLICATION_DEFENSE_BASELINES, type H6BenchmarkDataset, H6BenchmarkDatasetSchema, type H6TrapId, H6_ACTION_INTENT_JSON_SCHEMA, H6_ARMS, H6_DATASET_JSON_SCHEMA, H6_DECISION_RULE, H6_FROZEN_EVIDENCE_PATHS, H6_FROZEN_INVENTORY_HASH, H6_FROZEN_SEED, H6_FROZEN_SPLITS, H6_SUPPORT_ARTIFACT_PATHS, H6_TASK_JSON_SCHEMA, H6_TRAP_FINGERPRINT_JSON_SCHEMA, H6_TRAP_IDS, HOST_FAULT_RETRY_LIMIT, type HarnessRng, INJECTION_SUITE_ARMS, INJECTION_SUITE_FAMILIES, INJECTION_SUITE_PUBLICATION_ARMS, INJECTION_SUITE_STAGES, INJECTION_SUITE_VERSION, INTEGRITY_CIPHER_ALGORITHM, INTEGRITY_HASH_ALGORITHM, INTEGRITY_META_FIELDS, INVENTED_DOMAINS, type IngestionBenchAdapter, type IngestionLog, type InjectionSuiteArm, type InjectionSuiteCliInput, type InjectionSuiteCliResult, type InjectionSuiteEpisodeRow, type InjectionSuiteFamily, type InjectionSuiteOnlineAdaptiveInput, type InjectionSuiteProductEvidence, type InjectionSuiteRowIdentity, type InjectionSuiteStage, JUDGE_CALIBRATION_KAPPA_THRESHOLD, JUDGE_CALIBRATION_PROTOCOL_VERSION, type JudgeCalibrationCheckpointProvenance, type JudgeCalibrationIdentities, type JudgeCalibrationResult, type JudgeCategory, type KappaConfidenceInterval, LOCAL_LAB_PROVIDER_KINDS, LOCOMO_DATASET_FILENAMES, LOCOMO_FULL_TASK_COUNT, LOCOMO_RECALL_DIFF_LINE_LIMIT, LOCOMO_RECALL_EXCERPT_CHARS, LOCOMO_RETRIEVAL_TRACE_BUDGET_VERSION, LOCOMO_RETRIEVAL_TRACE_DELTA_SCHEMA_VERSION, LOCOMO_RETRIEVAL_TRACE_SCHEMA_VERSION, LOCOMO_RETRIEVAL_TRACE_SELECTION_VERSION, LONG_MEM_EVAL_DATASET_FILENAMES, type LeaderboardArtifactWrite, type LettaAdapterConfig, LettaMemCorrectAdapter, type LlmJudge, type LlmProvider, type LoCoMoCategory, type LoCoMoRetrievalMechanism, type LoCoMoRetrievalMechanismSummary, type LoCoMoRetrievalSessionReceipt, type LoCoMoRetrievalStructuralTrace, type LoCoMoRetrievalTaskDelta, type LoCoMoRetrievalTaskReceipt, type LoCoMoRetrievalTraceCoreCaptureReceipt, type LoCoMoRetrievalTraceDeltaReport, type LoCoMoRetrievalTraceProfile, type LoCoMoRetrievalTraceReceipt, type LoCoMoRetrievalTraceSelectionManifest, type LoCoMoRetrievalTraceSelector, type LoCoMoStructuralMultisetDelta, type LoComoCategoryDelta, type LoComoFinalContextRegression, type LoComoMetricDelta, type LoComoProfileArtifactEvidence, type LoComoProfileDeltaReport, type LoComoRawResultEvidence, type LoComoRecallCategoryDelta, type LoComoRecallContextSummary, type LoComoRecallDeltaReport, type LoComoRecallLineDelta, type LoComoRecallLineEvidence, type LoComoRecallMetricDelta, type LoComoRecallResultProvenance, type LoComoRecallTextDigest, type LoComoTaskRegression, type LoadDatasetOptions, type LoadSealedQrelsOptions, type LoadedDataset, type LoadedJudgeCalibrationState, type LocalLabManifest, type LocalLabManifestNotes, type LocalLabPhase, type LocalLabPhaseDescriptor, type LocalLabPhaseExecute, type LocalLabPhaseName, type LocalLabPhaseOutcome, LocalLabPreflightError, type LocalLabPreflightFailure, type LocalLabPreflightInput, type LocalLabPreflightOptions, type LocalLabPreflightResult, type LocalLabPreflightSuccess, type LocalLabProviderKind, type LocalLabRoleConfig, type LocalLlmProviderConfig, MAX_ROW_ATTEMPTS, MEMCORRECT_CORRECTION_ACCEPTANCE_RUBRIC, MEMCORRECT_CORRECTION_ACCEPTANCE_RUBRIC_VERSION, MEMCORRECT_STALE_HARM_RUBRIC, MEMCORRECT_STALE_HARM_RUBRIC_VERSION, MEMORY_EVAL_DIMENSIONS, MEMORY_EVAL_PUBLIC_LINE, MIN_CALIBRATION_SOURCE_TASKS, MITIGATED_BASELINE_SCENARIOS, type MaterializeOptions, type MaterializedRepo, type McpArgumentSemantic, type McpBackendErrorCode, type McpBackendResult, type McpBenchMemoryAdapter, type McpConformanceResult, type McpHttpTransportConfig, type McpListedTool, type McpMemCorrectAdapter, type McpMemoryAdapterOptions, McpMemoryBackendError, type McpMemoryToolMapping, type McpMemoryTransportConfig, type McpStdioTransportConfig, type McpToolCallResult, type McpToolClient, type McpToolMappingEntry, type McpToolMappingValue, type McpToolOperation, type Mem0AdapterConfig, Mem0MemCorrectAdapter, type MemCorrectGeneratorOptions, type MemCorrectJudgeRequest, type MemCorrectJudgeResult, type MemCorrectSystemAdapter, type MemoryEvalCategory, type MemoryEvalDimension, type MemoryEvalDimensionId, type MemoryEvalMetric, type MemoryGraph, type MemoryStats, type MemorySystem, type Message, type MetricAggregate, type MicroMetric, MissingCredentialError, type MitigatedBaselineConfig, type MitigatedTargetConfig, type MultipleChoiceQuestion, ONLINE_ADAPTIVE_STAGE, OPENAI_RESPONSES_JUDGE_RUBRIC_VERSION, OTHER_NAMESPACE_MEMORIES, type OllamaChatMessage, type OllamaProviderConfig, type OnlineAdaptiveCorpusLine, type OnlineAdaptiveCorpusManifest, type OnlineAdaptiveStatistics, type OpenAiCompatibleProviderConfig, OpenAiResponsesJudgeError, type OpenAiResponsesJudgeErrorCode, type OpenAiResponsesJudgeTelemetry, OpenAiResponsesProvider, type OpenAiResponsesProviderConfig, type OpenAiResponsesVerdict, type OpenAiResponsesVerdictResult, PROCEDURAL_REAL_SCENARIOS, PROCEDURAL_REAL_SCENARIOS_SMOKE, PUBLISHED_BENCHMARK_ARTIFACT_IDS, type PairedAnswerReplayCache, type PairedAnswerReplayEntry, type ParsedOllamaChatResponse, type PersonalizationRetrievalCase, type PreflightDiscoveredModel, type ProceduralAblationArtifact, type ProceduralAblationPerCase, type ProceduralAblationScenario, type ProceduralRealScenario, type ProceduralRealScenarioCategory, type ProviderBaseConfig, type ProviderConfig, type ProviderDiscoveryResult, type ProviderFactoryConfig, type PublishSkipReason, type PublishSkipRecord, type PublishedBenchmarkFeed, type PublishedBenchmarkFeedEntry, type PublishedBenchmarkId, REPEATED_FAILURE_ARMS, REPEATED_FAILURE_CONFIDENCE_LEVEL, REPEATED_FAILURE_INVALID_REASONS, REPEATED_FAILURE_STATISTICS_DRAWS, REQUIRED_FRONTMATTER_FIELDS, type RecallMetrics, type RecoveredMemory, type RegressionDetail, type RegressionGateResult$1 as RegressionGateResult, type RemnicAdapterOptions, type RepeatedFailureActionEvaluator, type RepeatedFailureArm, type RepeatedFailureCheckpointLoadResult, type RepeatedFailureCliCommandResult, type RepeatedFailureEffectAnalysis, type RepeatedFailureEpisode, type RepeatedFailureEpisodeDriver, type RepeatedFailureEpisodeEvidence, type RepeatedFailureEpisodeInput, type RepeatedFailureEpisodeRow, type RepeatedFailureExpectedDesign, type RepeatedFailureFactPairAudit, type RepeatedFailureFinalRepoEvidence, type RepeatedFailureFinalState, type RepeatedFailureGateEvent, type RepeatedFailureHolmResult, type RepeatedFailureInterval, type RepeatedFailureInvalidReason, type RepeatedFailureIsolationIdentity, type RepeatedFailureLocalToolHost, type RepeatedFailureNullableInterval, RepeatedFailureOllamaChatDriver, type RepeatedFailureOllamaChatDriverConfig, type RepeatedFailureProposedAction, type RepeatedFailureRowCheckpoint, type RepeatedFailureRowClaim, type RepeatedFailureRowIdentity, RepeatedFailureRowStore, type RepeatedFailureRowStoreOptions, type RepeatedFailureRunMetadata, type RepeatedFailureStatisticalAnalysis, type RepeatedFailureSuiteManifest, type RepeatedFailureSupportDecision, type RepeatedFailureTaskCut, type RepeatedFailureTimidityAnalysis, type RepeatedFailureTokenUsage, type RepeatedFailureTokenizer, type RepeatedFailureToolDefinition, type RepeatedFailureToolExecutionResult, type RepeatedFailureTrapAuditArtifact, type RepeatedFailureTrapAuditExpected, type RepeatedFailureTrapAuditMetrics, type RepeatedFailureTrapAuditRow, type RepeatedFailureTrapAuditRowIdentity, type RepeatedFailureTrapAuditThresholds, type RepeatedFailureTry, type ReplayRepeatedFailureStatisticsOptions, type ReportCardProvenanceContext, type ResolveBenchRuntimeProfileOptions, type ResolvedBenchRuntimeProfile, type ResolvedLocalLabProfile, type ResolvedLocalLabRole, type ResolvedRunBenchmarkOptions, type ResponsesApiOutputItem, type ResponsesApiRequest, type ResponsesApiResponse, type ResponsesApiUsage, type RetrievalMissStage, type RevisionShas, type RotatedChoices, type RunBenchmarkOptions, type RunJudgeCalibrationOptions, type RunProceduralAblationCliArgs, type RunProceduralAblationOptions, type RunRepeatedFailureCliCommandInput, type RunRepeatedFailureSuiteOptions, type RunRepeatedFailureSuiteResult, type RunSequentialPhasesOptions, type RunTrapAuditOptions, SCHEMA_TIER_FIXTURE, SCHEMA_TIER_SMOKE_FIXTURE, SEALED_PROMPT_REGISTRY, SINGLE_FLAG_ABLATION_MATRIX, STATE_CLASSIFICATIONS, SYNTHETIC_MEMORIES, type SanitizedDiagnosticProvider, type SavedBaseline, type SchemaTierCorpus, type SchemaTierFixture, type SchemaTierName, type SchemaTierPage, type SchemaTierPageFrontmatter, type SealedArtifact, type SealedJudgeDecision, type SealedJudgeInput, type SealedQrelsArtifact, type SealedQrelsHandle, type SealedRubric, type SearchResult, type SeededMemory, type SeededRandom, type SeededRng, type SequentialPhaseHooks, type SingleFlagAblationCell, type SingleFlagAblationId, type SpotCheckLogger, type StageObservation, type StageStatus, type StateClassification, type StateEvaluationResult, type StatisticalReport, type StrategyPatch, StrategyPatchSchema, type StructuredJudge, StructuredJudgeError, type StructuredJudgeErrorCode, type StructuredJudgeProvider, type StructuredJudgeTelemetry, type StructuredJudgeVerdict, type StructuredJudgeVerdictResult, type StructuredVerdictRequest, type SyntheticEdge, type SyntheticEmailIngestionAdapterOptions, type SyntheticFile, type SyntheticFileIR, SyntheticFileSchema, type SyntheticRepoConfig, type SyntheticSymbol, type SyntheticTargetOptions, TRAP_TAXONOMY, type TaskAttribution, type TaskAttributionGoldWitnessV1, type TaskAttributionRetrievalWitnessV1, type TaskAttributionWitness, type TaskAttributionWitnessRuntimeV1, type TaskAttributionWitnessV1, type TaskResult, type TaskTokenUsage, type TaskVariant, TaskVariantSchema, type TemporalRetrievalCase, type ThirdPartyAdapterConfig, type TierDetail, type TimelineEntry, type TokenUsage, type TrapFingerprintV1, TrapFingerprintV1Schema, type TrapTaxonomyItem, TrapTaxonomyItemSchema, type ValidationIssue, type ValidationReport, type WallMetric, type WriteBenchmarkArtifactResult, type ZepAdapterConfig, ZepMemCorrectAdapter, addContaminationEntry, aggregateTaskScores, analyzeInjectionSuiteOnlineAdaptiveRun, analyzeInjectionSuitePublicationRows, analyzeInjectionSuitePublicationRun, analyzeInjectionSuitePublicationUtility, analyzeInjectionSuitePublicationUtilityFile, analyzeInjectionSuiteRun, analyzeInjectionSuiteUtility, analyzeRepeatedFailureRows, answerBenchmarkQuestion, applyPatchAndCommit, assertCanaryUnderFloor, assertIntegrityMetaPresent, assertPublishableIntegrity, assertSha256Hex, assertTrapDatasetPreflight, assistantMeetingPrepDefinition, assistantMorningBriefDefinition, assistantNextBestActionDefinition, assistantSynthesisDefinition, attributeGoldMemory, attributeRun, attributeTask, backlinkF1, binarizeJudgeScore, bootstrapCohensKappaConfidenceInterval, bootstrapMeanConfidenceInterval, buildAmaBenchDiagnosticMatrixArtifact, buildAmaBenchDiagnosticVariantSummary, buildAmaBenchLeaderboardRows, buildBaselineFromReport, buildBenchmarkArtifact, buildBenchmarkArtifactFilename, buildBenchmarkPublishFeed, buildBenchmarkReproManifest, buildBenchmarkRunSeeds, buildBuildWeekEvidenceReceipt, buildCodexCreditReceipt, buildDriftCorpus, buildJudgePayload, buildOracleTrajectoryRecall, buildProviderFreeLoCoMoRetrievalConfig, buildRepeatedFailureRowKey, buildSchemaTierFixture, buildSchemaTierSmokeFixture, calculateCodexBudgetUnits, calculateJaccardSimilarity, calendarFixture, canonicalJsonStringify, captureBenchmarkExecutionProvenance, captureLoCoMoRetrievalTrace, captureMachineFingerprint, chatFixture, checkCodingGraphRegression, checkDatasetContamination, checkRegression, clampScore, cohensD, compareResults, computeBenchmarkReproDatasetInventoryHash, computeBenchmarkReproManifestArtifactHash, computeCohensKappa, computeH6InventoryHash, computeH6SupportArtifactHashes, computeRevisionShas, computeSealHash, computeTrapAuditArtifactHash, computeTrapAuditMetrics, containsAnswer, createSeededRng$1 as createAdamSeededRng, createAmaBenchDiagnosticAdapter, createAnthropicProvider, createCanaryAdapter, createClaudeCliProvider, createCodexCliProvider, createSeededRng as createCodingGraphSeededRng, createControlledResponsesAgentDriver, createDeterministicSpotCheckLogger, createGatewayResponder, createLightweightAdapter, createLiteLlmProvider, createLocalLlmProvider, createMcpDemoMemCorrectAdapter, createMcpDemoMemoryAdapter, createMcpMemCorrectAdapter, createMcpMemoryAdapter, createMitigatedTarget, createOllamaProvider, createOpenAiCompatibleProvider, createOpenAiResponsesBenchJudge, createOpenAiResponsesProvider, createSeededRandom$1 as createProceduralAblationSeededRandom, createProvider, createProviderBackedAmaBenchRecommendedJudge, createProviderBackedJudge, createProviderBackedResponder, createProviderBackedStructuredJudge, createRemnicAdapter, createRepeatedFailureOllamaChatDriver, createResponderFromProvider, createSeededRandom, createSeededRng$2 as createSeededRng, createSpotCheckFileLogger, createStructuredBenchJudge, createStructuredJudgeFromProvider, createSyntheticEmailIngestionAdapter, createSyntheticTarget, createTimeoutGuardedAdapter, decideInjectionSuiteCampaign, decideRepeatedFailureContent, decideRepeatedFailureStudy, decideRepeatedFailureTiming, defaultBenchmarkBaselineDir, defaultBenchmarkPublishPath, deleteBenchmarkResults, diagnoseLoCoMoRetrievalTraceDelta, diagnoseLoComoProfileDelta, diagnoseLoComoRecallDelta, discoverAllProviders, discoveryEndpointFor, emailFixture, entityRecall, evaluateTaskState, exactMatch, executeLocalRow, extractMetrics as extractCodingGraphMetrics, extractContentWords, extractMarkdownSectionsByTitle, f1Score, fixtureToAblationScenarios, formatHandoffNote, formatMissingDatasetError, generateDriftCorpus, generateFamilyVariants, generateH6BenchmarkDataset, generateReport, generateSuiteVariants, generateSyntheticRepo, getAblationCell, getBenchmark, getBenchmarkLowerIsBetter, getGitSha, getMemoryEvalDimension, getProviderBackedJudgePromptIdentity, getRemnicVersion, getTrapTaxonomyItem, hashBenchmarkArtifact, hashBytes, hashCanonicalJson, hashOrderedQuestionIds, hashString, holmAdjust, injectionSuiteResumeContractHash, integrityMetaIsComplete, interpretEffectSize, isAmaBenchUnknownLikeAnswer, isContaminationEntry, isContaminationManifest, isFrozenEvidencePath, isRepeatedFailureTimidityEquivalent, isSafeSyntheticPath, isSealedQrelsArtifact, isSha256Hex, isStructuredJudgeProvider, isTaskFailed, judgeMemCorrectCorrectionAcceptance, judgeMemCorrectStaleMemoryHarm, lexicalSimilarity, linkMatches, listBenchmarkBaselines, listBenchmarkResults, listBenchmarks, listMemoryEvalBenchmarkIds, listMemoryEvalDimensions, llmJudgeScore, llmJudgeScoreDetailed, loadAblationFixture, loadBaseline, loadBeamDatasetPreview, loadBenchmarkArtifact, loadBenchmarkBaseline, loadBenchmarkReportCardProvenance, loadBenchmarkResult, loadBenchmarkResultSummaries, loadCommittedH6BenchmarkDataset, loadCustomBenchmarkFile, loadJudgeCalibrationState, loadLoCoMo10, loadLocalLabManifest, loadLongMemEvalS, loadSealKeyFromEnv, loadSealedQrels, loadSealedRubric, matchEntity, materializeTaskRepo, mergeContaminationManifests, openSeal, orchestrateBenchmarkRuns, pairedDeltaConfidenceInterval, parseBenchmarkArtifact, parseCanaryFloor, parseCodexJsonlUsage, parseCustomBenchmark, parseLocalLabManifest, parseRepeatedFailureEpisodeRow, parseRubricResponse, parseSealedQrels, pickOne, pickStableQualifiedName, planInjectionSuiteRows, planOnlineAdaptiveRows, precisionAtK, preflightLoCoMoRetrievalTraceCapture, preflightLocalLabRole, projectFolderFixture, randomInt, recallAtK, reconcileCodexCreditLedger, redactBenchmarkResultSecrets, relativeRiskReduction, renderAttributionReportTable, renderBaselineMarkdown, renderBenchmarkResultExport, renderLoComoProfileDeltaMarkdown, renderLoComoRecallDeltaMarkdown, renderMemorySummaryForJudge, renderMemoryViewForAgent, replayInjectionSuiteStatistics, replayRepeatedFailureStatistics, resolveAssistantAgent, resolveAssistantRubricId, resolveAssistantSeeds, resolveAssistantSpotCheckDir, resolveBenchRuntimeProfile, resolveBenchmarkPhaseTimeoutMs, resolveBenchmarkProgressLogging, resolveBenchmarkResultReference, resolveBenchmarkRunCount, resolveBenchmarkRunId, resolveCanaryFloorFromEnv, resolveCodexCreditBudgetConfig, resolveCommittedH6FixtureDirectory, resolveEffectiveCanaryFloor, resolveLocalLabJudgeProviderConfig, resolveLocalLabProfile, resolveLocalLabRole, resolveStructuredJudge, rotateDistractors, rougeL, runAssistantBenchmark, runAssistantMeetingPrepBenchmark, runAssistantMorningBriefBenchmark, runAssistantNextBestActionBenchmark, runAssistantSynthesisBenchmark, runAttributeCliCommand, runBaseline, runBenchSuite, runBenchmark, runCodingGraphBenchmark, runCustomBenchmarkFile, runDriftGenCliCommand, runExplain, runExtractionAttack, runInjectionSuiteCliCommand, runInjectionSuiteOnlineAdaptive, runInjectionSuiteUtility, runJudgeCalibration, runMitigatedBaseline, runProceduralAblation, runProceduralAblationCli, runRepeatedFailureCliCommand, runRepeatedFailurePaperReportCliCommand, runRepeatedFailureSuite, runSealedJudge, runSequentialPhases, runTrapAudit, runTrapAuditCliCommand, runWithinCodexCreditBudget, safeHexEqual, sanitizeBenchmarkResultForJson, sanitizeLoComoResultReference, saveBaseline, saveBenchmarkBaseline, schemaCompleteness, sealPayload, selectAmaBenchDiagnosticVariants, selectCalibrationSlice, selectFixtureVariant, serializeAttributionReport, serializeBenchmarkArtifact, serializeBuildWeekEvidenceReceipt, serializeH6FixtureJson, serializeJsonl, serializeLoCoMoRetrievalTraceDelta, serializeLoCoMoRetrievalTraceReceipt, serializeSealedQrels, shuffleTasks, shuffled, stripCodeFences, summarizeBenchmarkResult, timed, tokenizeContent, unresolvedHelperImports, validateDriftCorpus, validateH6Dataset, validateH6FixtureBundle, validateH6StateDefiningIndependence, validateOllamaChatEndpoint, verifyMatchingTrapAudit, verifyRubricDigest, verifyTrapAuditArtifact, writeBenchmarkArtifact, writeBenchmarkPublishFeed, writeBenchmarkReproManifest, writeBenchmarkResult, writeBuildWeekEvidenceReceipt, writeH6FixtureBundle, writeJudgeCalibrationState, writeLeaderboardArtifactsForResult, writeRepeatedFailurePaperArtifacts, writeRepeatedFailureRunMetadata, writeRepeatedFailureStatistics, zeroScores };

package/dist/index.js

@@ -22,6 +22,7 @@ import {
   H6_ARMS,
   H6_DATASET_JSON_SCHEMA,
   H6_DECISION_RULE,
+  H6_FROZEN_EVIDENCE_PATHS,
   H6_FROZEN_INVENTORY_HASH,
   H6_FROZEN_SEED,
   H6_FROZEN_SPLITS,
@@ -113,6 +114,7 @@ import {
   isBenchmarkRunBlockedError,
   isContaminationEntry,
   isContaminationManifest,
+  isFrozenEvidencePath,
   isRepeatedFailureTimidityEquivalent,
   isSafeSyntheticPath,
   isSecretKey,
@@ -188,7 +190,7 @@ import {
   writeLeaderboardArtifactsForResult,
   writeRepeatedFailureRunMetadata,
   writeRepeatedFailureStatistics
-} from "./chunk-P4VDWOOT.js";
+} from "./chunk-LJCLTHYJ.js";
 
 // src/build-week-evidence-receipt.ts
 import { createHash as createHash2 } from "crypto";
@@ -53530,7 +53532,7 @@ async function writeRepeatedFailurePaperArtifacts(options) {
 }
 async function runRepeatedFailurePaperReportCliCommand(options) {
   try {
-    const { replayRepeatedFailureStatistics: replayRepeatedFailureStatistics2 } = await import("./repeated-failure-suite-runner-RVTM2OFY.js");
+    const { replayRepeatedFailureStatistics: replayRepeatedFailureStatistics2 } = await import("./repeated-failure-suite-runner-TPK5FQXU.js");
     const replay = await replayRepeatedFailureStatistics2(options);
     if (replay.exitCode !== 0) return replay;
     const result = await writeRepeatedFailurePaperArtifacts(options);
@@ -53859,6 +53861,7 @@ export {
   H6_ARMS,
   H6_DATASET_JSON_SCHEMA,
   H6_DECISION_RULE,
+  H6_FROZEN_EVIDENCE_PATHS,
   H6_FROZEN_INVENTORY_HASH,
   H6_FROZEN_SEED,
   H6_FROZEN_SPLITS,
@@ -54090,6 +54093,7 @@ export {
   isAmaBenchUnknownLikeAnswer,
   isContaminationEntry,
   isContaminationManifest,
+  isFrozenEvidencePath,
   isRepeatedFailureTimidityEquivalent,
   isSafeSyntheticPath,
   isSealedQrelsArtifact,

package/dist/{repeated-failure-suite-runner-RVTM2OFY.js → repeated-failure-suite-runner-TPK5FQXU.js}

@@ -11,7 +11,7 @@ import {
   runRepeatedFailureCliCommand,
   runRepeatedFailureSuite,
   verifyPreregistrationBinding
-} from "./chunk-P4VDWOOT.js";
+} from "./chunk-LJCLTHYJ.js";
 export {
   baseEvidence,
   buildDesign,

package/fixtures/h6-failure-gate/decision-rule-timing.json

@@ -0,0 +1,131 @@
+{
+  "version": 13,
+  "name": "H6 Timing Confirmatory Rerun Decision Rule",
+  "preregistration": {
+    "path": "docs/research/failure-gate/preregistration-timing-rerun.md",
+    "sha256": "4cf775705060d53f37fdc993f5a081d3b68b1646b47f9091889f4ecf2e4b2676"
+  },
+  "design": {
+    "armsMode": "timing_only",
+    "arms": [
+      "TURN_START_FAILURE",
+      "PRE_ACTION_FAILURE"
+    ]
+  },
+  "analysisPopulation": {
+    "datasetVersion": 1,
+    "datasetInventoryHash": "687615b5f7ff46977d268a03e30018070f7d0bec9d01e04da2d0c723e59a5b27",
+    "split": "main",
+    "taskCount": 18,
+    "seedsPerModelProfile": 5,
+    "modelProfileCount": 1,
+    "unit": "task",
+    "pairingKey": [
+      "taskId",
+      "variantId",
+      "seed",
+      "modelProfileId",
+      "modelProfileHash"
+    ],
+    "maximumPrimaryTaskCuts": 0,
+    "registeredSeeds": [
+      6,
+      7,
+      8,
+      9,
+      10
+    ]
+  },
+  "analysis": {
+    "bootstrap": {
+      "draws": 10000,
+      "group": "task",
+      "confidenceLevel": 0.95,
+      "method": "percentile"
+    },
+    "shuffle": {
+      "draws": 10000,
+      "group": "task",
+      "alternative": "candidate_benefit",
+      "plusOneCorrection": true
+    },
+    "alpha": 0.05,
+    "multiplicity": {
+      "method": "HOLM",
+      "family": [
+        "H6_TIMING_REPEATED_FAILURE"
+      ]
+    }
+  },
+  "invalidRowPolicy": {
+    "VAGUE_CHECK": "WORST_CASE_AGAINST_CANDIDATE"
+  },
+  "factMatching": {
+    "timing": {
+      "factCount": 1,
+      "requireSameFactIds": true,
+      "requireSameCitationHashes": true,
+      "requireSameRenderedTokenCount": true
+    }
+  },
+  "hypotheses": {
+    "H6-timing": {
+      "baselineArm": "TURN_START_FAILURE",
+      "candidateArm": "PRE_ACTION_FAILURE",
+      "metric": "repeatedFailure",
+      "minimumRelativeRiskReduction": 0.3,
+      "minimumAbsoluteRepeatedFailureBenefit": 0.05,
+      "requireRepeatedFailureBenefitIntervalLowerStrictlyAbove": 0,
+      "requireHolmAdjustedPStrictlyBelow": 0.05,
+      "zeroBaselineDecision": "NOT_ESTIMABLE"
+    }
+  },
+  "power": {
+    "simulationDraws": 10000,
+    "minimumTimingPower": 0.8,
+    "sourceSplit": "pilot",
+    "transferredPilotEvidence": {
+      "runId": "h6-a30b5cb7dc9174e31329195d",
+      "manifestArtifactHash": "5bce2c0c20fadf77706ae5dbb503d4f938d40e2d4178c5395b36683cfef75e21",
+      "powerArtifactHash": "cfaec22f374b353b536dec6067243a1731804f8f1d8d2f4742e1631cf1b27bd5",
+      "decisionRuleHash": "c775947440dc764a62fbc4bd0c3fca08fe2ede90b543701875af3ab541114cd9",
+      "preregistrationHash": "0fe2838c50d3e70be59b47bc5d932fecd52af8975c6c0279b3ebc0ebf4a80236",
+      "harnessSourceHash": "129045d73b7f8fdc18bb5b9a87faa5245f9921945ce1ff9d94d859f29bb88770",
+      "analysisVersion": "h6-task-bootstrap-shuffle-holm-v1",
+      "timingPower": 0.8364
+    }
+  },
+  "trapAudit": {
+    "minimumTrappedRate": 0.3,
+    "minimumNonFixedRate": 0.5,
+    "maximumInvalidRows": 0,
+    "requireCompleteRows": true
+  },
+  "completeness": {
+    "expectedRowsFormula": "taskCount*variantsPerTask*seedsPerModelProfile*modelProfileCount*armCount",
+    "primaryArmCount": 2,
+    "hostFaultRetriesAfterFirstTry": 5,
+    "rerunTaskResults": false,
+    "invalidReasons": [
+      "CORPUS_INVALID",
+      "CORE_REPO_DIR_MISMATCH",
+      "START_DRIFT",
+      "TRACE_GAP",
+      "VAGUE_CHECK",
+      "MIXED_ARM_STATE",
+      "UNMATCHED_FACTS",
+      "WAIT_RULE_FAULT",
+      "HOST_RETRIES_EXHAUSTED"
+    ]
+  },
+  "outcomes": {
+    "PASS": "timing=SUPPORTED",
+    "REJECT": "timing=REJECTED",
+    "NOT_ESTIMABLE": "timing=NOT_ESTIMABLE"
+  },
+  "gateStatuses": [
+    "NO_MATCH",
+    "MATCH_WARN",
+    "ERROR_FAIL_OPEN"
+  ]
+}

package/fixtures/h6-failure-gate/tasks/task_01_cache_invalidation/d0/README.md

@@ -0,0 +1,7 @@
+# Cache Expiration Invalidation
+
+This package contains the base implementation and proposed candidate refactoring strategies for Cache Expiration Invalidation.
+
+## Structure
+- `src/`: Source files for the module.
+- `TASK.md`: Specification of the required modification.

package/fixtures/h6-failure-gate/tasks/task_01_cache_invalidation/d0/TASK.md

@@ -0,0 +1,11 @@
+# Task: Cache Expiration Invalidation
+
+## Objective
+Update the CacheManager module to ensure expired entries are purged during read operations.
+
+## Available Strategies
+1. **candidate_alpha**: Check expiration during read and delete key from store if expired.
+2. **candidate_beta**: Extend TTL timer on access whenever entry is fetched.
+
+## Instructions
+Review the implementation under `src/` and select or apply the appropriate candidate modification.

package/fixtures/h6-failure-gate/tasks/task_01_cache_invalidation/d0/src/index.ts

@@ -0,0 +1,13 @@
+export class CacheManager<K, V> {
+  private store = new Map<K, { value: V; expiresAt: number }>();
+
+  public set(key: K, value: V, ttlMs: number): void {
+    this.store.set(key, { value, expiresAt: Date.now() + ttlMs });
+  }
+
+  public get(key: K): V | undefined {
+    const entry = this.store.get(key);
+    if (!entry) return undefined;
+    return entry.value;
+  }
+}

package/fixtures/h6-failure-gate/tasks/task_01_cache_invalidation/d0/src/utils.ts

@@ -0,0 +1,3 @@
+export function isExpired(expiresAt: number, now = Date.now()): boolean {
+  return now >= expiresAt;
+}

package/fixtures/h6-failure-gate/tasks/task_01_cache_invalidation/d1/README.md

@@ -0,0 +1,7 @@
+# Cache Expiration Invalidation
+
+This package contains the base implementation and proposed candidate refactoring strategies for Cache Expiration Invalidation.
+
+## Structure
+- `src/`: Source files for the module.
+- `TASK.md`: Specification of the required modification.

package/fixtures/h6-failure-gate/tasks/task_01_cache_invalidation/d1/TASK.md

@@ -0,0 +1,11 @@
+# Task: Cache Expiration Invalidation
+
+## Objective
+Update the CacheManager module to ensure expired entries are purged during read operations.
+
+## Available Strategies
+1. **candidate_alpha**: Check expiration during read and delete key from store if expired.
+2. **candidate_beta**: Extend TTL timer on access whenever entry is fetched.
+
+## Instructions
+Review the implementation under `src/` and select or apply the appropriate candidate modification.

package/fixtures/h6-failure-gate/tasks/task_01_cache_invalidation/d1/src/index.ts

@@ -0,0 +1,13 @@
+export class CacheManager<K, V> {
+  private entryRepository = new Map<K, { value: V; expiresAt: number }>();
+
+  public set(key: K, value: V, ttlMs: number): void {
+    this.entryRepository.set(key, { value, expiresAt: Date.now() + ttlMs });
+  }
+
+  public get(key: K): V | undefined {
+    const entry = this.entryRepository.get(key);
+    if (!entry) return undefined;
+    return entry.value;
+  }
+}

package/fixtures/h6-failure-gate/tasks/task_01_cache_invalidation/d1/src/utils.ts

@@ -0,0 +1,3 @@
+export function isExpired(expiresAt: number, now = Date.now()): boolean {
+  return now >= expiresAt;
+}

package/fixtures/h6-failure-gate/tasks/task_01_cache_invalidation/d2/README.md

@@ -0,0 +1,7 @@
+# Cache Expiration Invalidation
+
+This package contains the base implementation and proposed candidate refactoring strategies for Cache Expiration Invalidation.
+
+## Structure
+- `src/`: Source files for the module.
+- `TASK.md`: Specification of the required modification.

package/fixtures/h6-failure-gate/tasks/task_01_cache_invalidation/d2/TASK.md

@@ -0,0 +1,11 @@
+# Task: Cache Expiration Invalidation
+
+## Objective
+Update the CacheManager module to ensure expired entries are purged during read operations.
+
+## Available Strategies
+1. **candidate_alpha**: Check expiration during read and delete key from store if expired.
+2. **candidate_beta**: Extend TTL timer on access whenever entry is fetched.
+
+## Instructions
+Review the implementation under `src/` and select or apply the appropriate candidate modification.